基于异质网络的药物协同模式挖掘关键技术研究

CORC > 自动化研究所 > 中国科学院自动化研究所 > 毕业生 > 博士学位论文

题名	基于异质网络的药物协同模式挖掘关键技术研究
作者	陈迪
学位类别	工学博士
答辩日期	2016-05
授予单位	中国科学院大学
授予地点	北京
导师	杨一平
关键词	异质网络，药物协同，中药方剂，并发疾病，谱聚类，随机森林，随机游走，网络模体
中文摘要	药物协同是指不同药物相互合作以增强疗效的一种相互作用关系，在药物研发领域内具有重要的研究价值。癌症、艾滋病等复杂疾病通常由多种因素导致，涉及到多个生物环节的异常或改变。单一药物很难应对复杂疾病的系统性病变特点，而具有协同作用的药效成分组合，例如组合药物和中药方剂，则能够系统性干预到人体内多个环节实现对复杂疾病的治疗。然而，随着候选药效成分数目的增加，通过实验的方法从海量的候选组合中筛选出协同组合不仅效率低下、浪费资源，而且无法揭示潜在的协同作用机制。因此，如何利用现有的生物医学数据资源对药物协同进行有效预测并揭示潜在的协同作用机制是药物协同研究中的重要问题。本文充分考虑到复杂疾病相关要素的系统性、整体性，通过异质网络对多种数据资源进行整合利用，从而全面描述药物与人体间的复杂交互作用，并提出基于异质网络的药物协同分析与预测方法，利用关联分析、分类、聚类以及子图挖掘等技术方法，从组合药物、中药方剂、药物交互作用、并发疾病多个角度进行药物协同模式挖掘的相关研究，揭示具有协同作用的药物之间共同的关联机制和作用机理。本文的研究内容包括如下四个方面：（1）为了预测组合药物中两个药物间的协同作用，提出一种基于“蛋白质-通路”异质网络的预测方法。该方法利用现有的蛋白质、通路数据资源构建异质网络，并基于该异质网络的拓扑特征从通路间的关联关系和通路内部的蛋白质交互作用两个角度定义协同关联得分。对比于现有的网络药理学方法，该方法显著提高了预测精度。同时，将组合药物的靶标映射到此异质网络上还能够为揭示药物协同机制提供重要参考。（2）为了分析中药方剂中多种成分间的协同作用机理，提出了基于异质网络的中药方剂协同作用机理挖掘方法。该方法在分析方剂中每两个成分的协同关联性的基础上对方剂的组成成分进行聚类分析，并构建特定疾病相关的通路网络以及成分-通路-生物标记物网络，描述每个协同成分聚类的具体功能，从多角度揭示中药方剂的协同作用机制。本文将上述分析方法用于金柴抗病毒胶囊，发现该方剂协同作用于彼此间具有密切关系并且在流感病毒生命周期中发挥重要作用的多种通路。（3）为了对协同、拮抗、独立三类药物交互进行分类，提出一种基于多维关联特征的分类方法。该方法整合利用不同药物在化合物结构、ATC编码、靶标、酶、通路、基因本体等多个层面共123维的关联特征作为药物交互对的特征描述，并利用随机森林算法训练分类器，构建针对三类药物交互作用的分类模型。该方法分类精度较高、可解释性强，能够促进药物研发领域对药物交互作用的区分及对潜在机理的阐释。（4）为了揭示临床上针对多种并发疾病的协同用药规律，提出一种基于电子病历的并发疾病组合用药模式挖掘方法。该方法同时利用卡方检验和相对危险系数两种方法从电子病历中识别具有显著关联性的疾病-疾病、药物-药物、疾病-药物对，构建疾病-药物异质网络。在此异质网络的基础上，结合随机游走和网络模体分析方法识别其中反映并发疾病用药特点的关键网络模式。该方法能够有效揭示临床上针对并发疾病的协同用药规律，辅助提高对并发疾病的综合疗效。
英文摘要	Drug synergy refers to the interplay between different drugs, which can enhance drugs’ overall therapeutic effects. Drug synergy is of great value in the field of drug research and development. Complex diseases like cancer and AIDs are caused by multiple factors, with various biological processes being disturbed or changed. Single drug is insufficient to handle the systematic pathology. In contrast, combination drugs or TCM formulae, both of which are composed of multiple efficacy components with synergistic effects, can treat the complex diseases by cooperately acting on multiple pathological processes. With the number of candidate components increases, the number of all possible combinations will grow exponentially, it will be both time and resource-consuming to screen synergistic interactions by experimental methods. Besides, the synergy mechanism has not been fully defined. Consequently, how to predict synergistic effects and illustrate the mechanism based on currently available biomedical data resources has become a matter of concern. Considering the systematic and integral features of complex diseases, the paper integrated various kinds of data resources by heterogeneous network to describe the complicated interactions between drugs and human body’s bio-system. Based on these heterogeneous networks, drug synergy patterns were explored by ranking, classification, clustering and sub-graph mining algorithms, from different perspectives, including combination drugs, TCM formulae, drug interactions and comorbidities. More specifically, the main content of this paper has been concluded as bellow: (1) A “protein-pathway” heterogeneous network based method was proposed to predict the synergistic effects of combination drugs. This method integrated protein and pathway databases to build a “protein-pathway” heterogeneous network, based on which a synergy score was defined to evaluate the association between different drugs considering both pathway dependencies and protein interactions within individual pathways. Experiments show that this method could achieve better performance than pre-existing network pharmacology methods. In addition, by extracting the subgraph of synergistic combination drugs on the heterogeneous network, this method can offer meaningful mechanistic hypotheses on drug synergy. (2) A TCM formula is composed of numerous ingredients. To uncover the synergistic mechanism of these ingredients, this paper put forward a heterogeneous biology network based method. The synergy score between each two ingredients was calculated based on the “protein-pathway” heterogeneous network. Then all ingredients were clustered based on the scores, and specific disease-related pathway network and “ingredient-pathway-biomarker” network were built to further analyze the functions of each cluster, as well as to reveal the synergistic mechanism. This paper has applied such method on “Jin Chai” antiviral capsule to elucidate its synergistic effects. The results show that different ingredients cooperate with each other by targeting on multiple mutual-associated pathways which play crucial functions in the life cycle of influenza virus. (3) A classification model based on multi-dimensional association features was built to classify drug interactions into three types - synergistic, antagonistic and independent interactions. First, multi-dimensional association features were calculated considering the relationships of drugs on aspects of chemical structures, ATC codes, targets, enzymes, pathways, gene ontology and so on. Based on such features, a random forest classification model was trained to classify three types of drug interactions. This model is of good accuracy and explainability, it can not only achieve good accuracy, but also help explore the potential mechanism of drug interactions. (4) How to effectively combine different types of drugs in the treatment of comorbidity is of great importance in clinical practice. As to this problem, this paper put forward an electronic medical record (EMR) based method. First, chi-square test and relative risk coefficient were utilized to discover the significant associations among diseases and drugs from the EMRs, and a “disease-drug” heterogeneous network was constructed to describe the disease-disease, disease-drug and drug-drug associations. Then, combination medication patterns of comorbidity diseases, which can depict how different types of drugs can be combined to treat two comorbidity diseases, were mined from the heterogeneous network through random walk and network motif analysis algorithms. The method can reveal the medication principles for comorbidities and help improve treatment efficiency of comorbidities.
内容类型	学位论文
源URL	[http://ir.ia.ac.cn/handle/173211/11485]
专题	毕业生_博士学位论文
作者单位	中国科学院自动化研究所
推荐引用方式 GB/T 7714	陈迪. 基于异质网络的药物协同模式挖掘关键技术研究[D]. 北京. 中国科学院大学. 2016.